An Exploration of Social Identity: The Structure of the BBC News-Sharing Community on Twitter
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
An Exploration of Social Identity: The Structure of the BBC
News-Sharing Community on Twitter
Julius Adebayo, Tiziana Musso, Kawandeep Virdee, Casey Friedman, and Yaneer Bar-Yam
New England Complex Systems Institute
238 Main Street Suite 319, Cambridge, Massachusetts 02142, US
(Dated: August 20, 2013)
Abstract
Online social media influence the flow of news and other information, potentially altering col-
lective social action while generating a large volume of data useful to researchers. Mapping these
networks may make it possible to predict the course of social and political movements, technology
adoption, and economic behavior. Here we map the network formed by Twitter users sharing
British Broadcasting Corporation (BBC) articles. The global audience of the BBC is primarily
organized by language with the largest linguistic groups receiving news in English, Spanish, Rus-
sian and Arabic. Members of the network primarily “follow” members sharing articles in the same
language, and these audiences are primarily located in geographical regions where the languages
are native. The one exception to this rule is a cluster interested in Middle East news which includes
both Arabic and English speakers. We further analyze English-speaking users, which differentiate
themselves into four clusters: one interested in sports, two interested in United Kingdom (UK)
news—with word usage suggesting this reflects political polarization into Conservative and Labour
party leanings—and a fourth group that is the English speaking part of the group interested in
Middle East news. Unlike the previously studied New York Times news sharing network the largest
scale structure of the BBC network does not include a densely-connected group of globally inter-
ested and globally distributed users. The political polarization is similar to what was found for
liberal and conservative groups in the New York Times study. The observation of a primary orga-
nization of the BBC audience around languages is consistent with the BBC’s unique role in history
as an alternative source of local news in regions outside the UK where high quality uncensored
news was not available.
1
Tracing the flow of information in society can offer insights into the structure of social
groups–who associates with whom–and the patterns of response to various events [1, 2, 3,
4]. Here we map the structure of a network composed of individuals who share British
Broadcasting Corporation (BBC) news articles on the social networking website Twitter.
We find that the BBC global network separates into linguistically distinct regional groups.
The largest of these groups reads the BBC in English, with subgroups focused on sports,
UK news and business, among other topics. Analysis of two groups focused on UK news
suggests they are polarized politically into Labour and Conservative (Tory) party leaning
groups. Other major language groups include Spanish, Russian, and Arabic. The Arabic
speaking group is unique in being closely linked to an English speaking sub-group, together
focused on Middle East news. Headline articles in each language category typically pertain
to the corresponding region. These results differ from those we previously obtained for the
New York Times news-sharing network, which includes a substantial cosmopolitan group
that follows news from around the globe, in addition to subgroups focused on local and
national news [5]. The organization of the BBC network along linguistic lines is consistent
with provision of customized foreign language services and the historical prominence of the
BBC in regions where political constraints have inhibited the reliability of local news sources.
Social media platforms including Facebook and Twitter are providing an explosion of
information about interpersonal communication because these platforms provide a record
of ongoing of communication and response to information [3, 5, 6]. Increasingly, these
platforms serve as filters for, or determinants of, the information to which individuals are
exposed. As they interact on these platforms, individuals preferentially associate with others
from whom they want to receive information. These associations can be characterized as
links in a network, and the users as nodes. The substructure of the network can be identified
through the presence of densely interconnected components: users who are likely to reflect
common interests and other shared characteristics. Mapping network structure can therefore
elucidate the factors that drive social association, including shared interest, geography, age,
class, language, religion, race, or profession [7, 8, 9, 10]. Differences in the structure of
distinct news-sharing networks reflect differences in news coverage and the audience these
news sources attract. The ability to analyze social relationships and map the structure of
social communities enable an understanding of collective response to news events [11, 12].
Here we identify social groups among users sharing BBC articles on Twitter. The BBC
2is a semi-autonomous British public media corporation, with television, radio, and online
services. The BBC provides both United Kingdom (UK) and international news, as well as
other forms of programming. The BBC reaches 239 million people globally, and its news
coverage reaches over 80% of the UK adult population [13]. The international reach of the
BBC stems from the BBC World Service, founded in 1932 as a purely English-language
“Empire Service.” In 1938, the BBC began its first foreign language broadcasts, directed
at the Arab World and Latin America, in order to counter propaganda by European rivals
[14]. The run-up to war and World War II itself prompted the BBC to expand into many
other languages, including those of Nazi-occupied countries. Throughout its history, the
UK government, BBC staff, and historians have cast the World Service as both an objec-
tive news source and a public diplomacy tool to convey the British experience to far-flung
peoples [15, 16, 17]. Historical accounts suggest that the BBC gained global reputation for
reliability [14]. The BBC World Service has responded to changing conditions by initiating
or expanding broadcasts in languages spoken in places of geopolitical importance or national
crisis: introduction of a Latin American Service in 1938, German in 1939, Arabic after the
1956 Suez crisis, non-Russian languages of the former Soviet Union from 1991, and Albanian
in 1993 [17, 18].
BBC articles are currently being shared hundreds of thousands of times per week on
Twitter. Twitter is a micro-blogging platform through which individuals can publish and
read posts of up to 140 characters called “tweets” that may contain links to webpages
[19, 20]. Users subscribe to receive tweets from other users, indicating that they read their
tweets on a regular basis. This is called following. The follower relationship among users
identifies links that comprise a social network. We can characterize users by the subjects
of the articles they tweet. Each BBC Online article belongs to a subject category, which is
the first sub-directory in the URL of the article shared. For example, an article with the
URL: http://www.bbc.co.uk/sports/0/football/... belongs to the sports category. For our
analysis, we labeled users by their interest, inferred from the category of article they tweeted
the most. For additional dimensions along which to characterize users, we performed text
analysis of user provided fields: location and self-description. Identifying users by interest,
location, and self-description enables a characterization of emergent highly connected groups.
Methodological details are provided in the Appendix.
We collected tweets containing BBC online article links over a seven day period, March
3A# B#
C#
uk # # #(30.81%)# middle_east #(5.26%)#
mundo # #(18.14%)# health # #(4.24%)#
sport# # #(15.74%)# programmes #(4.16%)#
business # #(7.69%)# russian # #(4.16%)#
technology #(6.12%)# arabic# # #(3.65%)#
FIG. 1: (A) Network of Twitter users who share BBC online articles. Links are “follow” relation-
ships between users. The network layout is obtained by pulling followers and followees close to each
other, while pushing apart unconnected nodes. The long range links that are visible occur where
relatively few links connect groups of nodes that are otherwise pushed far apart. (B) Same as A
with links removed for clarity. (C) Same as B with user nodes colored according to the subject of
articles they share the most.
44 - 10, 2011, resulting in 489,878 tweets posted by 153,172 users. We considered only users
who shared at least three articles from a particular category resulting in 10,347 users. We
generated a network of BBC article-sharers by representing each Twitter user in our sample
as a node and each follower-followee relationship between users as an edge. The spatial
layout of the network, shown in Figure 1, is determined by a force directed layout algorithm,
which takes into account only the follow relationship between users. The layout algorithm
pulls together users (nodes) that are connected together and pushes apart unconnected
users, highlighting densely connected components of the overall network [21]. We performed
a clustering analysis that quantitatively separates clusters into distinct subgroups [5, 22].
The clustering of the network further manifests, visually, the presence of distinct highly
connected subcomponents.
Figure 1 shows the BBC news-sharer network with user nodes colored by interest. This
labeling enables us to characterize the composition of the clusters. Topic colors are not
randomly scattered; for part of the network, nodes of the same colors tend to be more
closely associated spatially. We identified seven constituent cluster communities. Figure 2
shows the result of the characterization of these clusters. The seven clusters can be classified
into four different groups primarily distinguished by language: English, Spanish, Russian,
and Middle East. The latter includes Arabic- and English-speaking parts. We summarize
the clusters found and their characterization as follows:
• The ‘English group’ consists of several clusters with users who primarily share English
articles. About 70 percent of all users share English language articles. The English
group further subdivides into four clusters, one focused on Sports and three focused
on combinations of UK and business news.
• The ‘Spanish group’ consists of users reading articles from the Spanish-language BBC
Mundo service. The BBC Mundo content in Spanish relates to Latin America and
Spain. These articles address regional issues such as economics, political instability,
and health care, with some global coverage as well. This cluster constitutes approx-
imately 18 percent of the users studied. About 65 percent of users in this group are
from Venezuela. Other countries represented include Chile, Mexico, and Spain.
• The ‘Russian group’ is primarily focused on BBC news articles in Russian. Articles
in Russian generally cover the state of the Russian economy and government, as well
5800 200 2000
600 150 1500
400 100 1000
200 50 500
0 0
0
no s
si y
uk
th
ss
es
pr usi uk
ra ss
rt
he t
or
ch e
bu log
o
th
gy
no k
ss
he t
al
ne
og ne
te mm
m
or
sp
sp
u
al
ne
lo
m
sp
ra
si
b
bu
og
ch
pr
300 te
300
200 200
100 100
0 0
gy
uk
n
no o
m rt
si h
he y
uk
ss
no t
ia
ch d
ch por
o
g
bu alt
lo
te un
sp
ne
ss
lo
s
ru
te
300 1500
200 1000
100 500
0 0
ic
si y
no k
ss
ar t
dl y
uk
no s
do
st
s
bu log
u
m log
ch s
ab
ea
ea
ne
te sine
un
dl
m
id
id
ch
bu
m
te
FIG. 2: Topics of interest: Article subjects most commonly shared by users in the clusters of Figure
1. The correspondence between the graphs and the clusters is by color.
as other international news. This cluster constitutes about 4 percent of the users
studied. The group has a sparse substructure of multiple subgroups that are not well
interconnected. Almost all users are from Russia, particularly Moscow. We also see
users from Ukraine and Karachi, Pakistan.
• The ‘Middle East group’ consists of two clusters and is the only mixed language
component. The first cluster is Arabic speaking while the other is English speaking.
The latter is particularly tightly linked. The dominant subject category is the Middle
6150 150 600
100 100 400
50 50 200
0
0 0
lo uk
en les
w n
r
ch d
te
o
an lan
nd
d
lo k
en on
sc nd
ki ited
uk
m
on
d
nd
es
a
u
ite
an
do
m g
la
nd
a
nd
un
un
gl
gl
ot
ng
lo
en
80 20
60 15
40 10
20 5
0 0
da
lo k
ca ia
al a
on
u
us
uk w
ka e
ka i
os a
n
s
ch
na
nd
in
i
za
ay
co
ss
ra
ra
ru
m
m
40 200
30 150
20 100
10 50
0
0
es ico
s
na
ca ela
m le
ca
uk
i
t
on
iro
t
ch
yp
ai
pa
ex
zu
ra
w
nd
ca
eg
ne
ku
lo
ve
FIG. 3: Geographic locations: Words most frequently contained in the user location field in the
clusters of the network in Figure 1. The correspondence between the graphs and the clusters is by
color.
East. Some users are in Arabic speaking countries such as Egypt and Kuwait, while
others are in the UK.
In order to clarify the substructure of the English speaking supercluster, we generated
a network consisting only of users who tweeted mostly English-language articles. Figure 5
illustrates the structure of connectedness and the distribution of individual interests within
this subnetwork. A clustering of the English subnetwork manifests four groups: one focused
7uk # # #(37.89%)# health # # # #(5.22%)#
sport # # #(19.39%)# programmes # # #(5.11%)#
business # #(9.46%)# science&environment# #(3.67%)#
technology# # #(7.52%)# entertainment&arts# # #(2.66%)#
middle_east# # #(6.47%)# europe# # # #(2.65%)#
FIG. 4: Network of Twitter users who tweet BBC online articles in English, constructed similarly
to Figure 1. User nodes are colored according to the topic of articles they share the most. Network
edges have been omitted for clarity.
on Sports, two on National News, and the fourth on the Middle East. The fourth group is
the English speaking part of the Middle East cluster previously identified. We summarize
the clusters found in Figure 5 as follows:
• Cluster A is primarily focused on sports with other secondary interests such as UK
news, business, programs, and technology. Users from this cluster are predominantly
from the UK.
8A
B
A
B
C
D
C
D
FIG. 5: Topics of interest: Article subjects most commonly shared by users in subgroups of the
English group of Figure 4. The correspondence between the graphs and the clusters is by color.
• Cluster B is primarily focused on UK news (news stories regarding England, Northern
Ireland, Scotland, and Wales) with secondary interests in business, technology, sports,
and science & environment. A significant percentage of users in this cluster are lo-
cated in the UK, with others indicating their location as United States, Canada, and
Malaysia–countries with strong historical ties to the UK. However, many countries
with large populations and strong British ties, such as India, are not represented.
• Cluster C is similar to Cluster B, though users exhibit a stronger topical focus on UK
news. Other secondary interests include business, sports, programs, and science &
environment. Users in this cluster are almost exclusively from the UK.
• Cluster D is focused on the Middle East. This is the English speaking portion of the
Middle East group indicated in Figure 1. The BBC Middle East service publishes
9international news stories about the Middle East region. A small percentage of users
are also interested in UK news, science & environment, Europe, and technology.
Clusters B and C show a remarkable similarity of subject categories. To further dis-
tinguish between them, we performed a text analysis of the users’ self-descriptions in each
cluster. Figure 6 shows the relative frequency of the words obtained above a threshold
frequency (0.007). Among the most commonly used words in Cluster B self-descriptions
are “business,” “marketing,” and “financial,” suggesting that members of Cluster B are
highly business-oriented. Among the common words from Cluster C are “politics,” “stu-
dent,” “science,” and “labour,” indicating that members of this group include politically
and technologically oriented individuals who may be supportive of the UK Labour Party.
We infer that these two clusters, B and C, may reflect polarization between two major sides
of the UK ideological political spectrum associated with the dominant political parties, the
Conservative party (Tory) and the Labour party respectively. This is similar to the liberal-
conservative political divide observed for the US, in a mapping of the New York Times news
sharing community on Twitter [5].
FIG. 6: User self-description: Words most frequently contained in user self-description fields in
sub-cluster B (left chart) and sub-cluster C (right chart).
Overall, the dominant structural determinants of the BBC article-sharing network are
geography and language. The BBC serves, for the most part, isolated regional audiences in-
10terested in regional news rather than a global audience interested in global affairs. Language,
and therefore location, tend to be important barriers between groups. The cluster structure
is likely to be rooted in the BBC’s history as an alternative local information source. One
implication of the close link between language and geography is that BBC Online’s Mundo
and Russian coverage primarily serves foreign populations rather than immigrant popula-
tions resident in the UK. The sharp division between regional sub-networks highlights the
BBC’s need to provide content tailored to different audiences.
From the BBC news sharer network shown in Figure 1, we observe that a handful of
countries–Chile, Egypt, Mexico, Pakistan, Russia, and Venezuela–account for most non-
English article sharing. A potential explanation for pockets of BBC influence abroad is a
historical absence of trustworthy or high-quality domestic news sources in these countries
that led their inhabitants to seek regional news from an outside news service to which they
have limited cultural ties. Indeed, there is a correspondence between these countries with
historical periods of lower press freedom, high values (less free) on the world press freedom
index, and the volume of article sharing in our sample. Chile, Egypt, Mexico, Pakistan,
Russia, and Venezuela all have had notable limitations to press freedom in the past justifying
a reliance on other news sources. There exist countries that have historically weak press
freedoms but few tweets in the sample studied. In some of these cases governments censor
news media and block social media. For example, even though the BBC has Chinese and
Farsi language services, there is little sharing of articles in these languages in our sample—
this can be readily understood because the governments of mainland China and Iran both
block public Twitter access.
In summary, access to new social media platforms and the increased propagation of news
has led to the emergence of new ways for individuals to organize, and created new channels
for coordinating group collective action [5]. Knowledge of cluster and user traits can enable a
better understanding of social dynamics and opportunities for news providing organizations.
Insights gained from our work on BBC news sharing during a particular window of time
reflects basic understanding about the BBC news organization and its audience globally. In
recent years, financial pressures have led to closure of some of the BBC foreign language
services [23, 24]. Information about the social networks formed may be helpful in developing
corporate strategies.
11Appendix
During March 4-10, 2011, we collected tweets that contain a URL from the “bbc.co.uk”
(including redirects from “bbc.in” and other top level domains) domain, using the Twit-
ter Application Programming Interface (API), resulting in 489,878 tweets posted by
153,172 unique users. Each user was labeled with a particular topic of interest cor-
responding to the specific genre of the articles tweeted. The topic genre is identified
from the URL of the article shared. For example, a user sharing an article with URL:
http://www.bbc.co.uk/sports/0/football/... would be in the sports category. As of March
2011, BBC had more than 20 main categories including sports, UK News, Mundo, and Rus-
sian. We focused on the 10 most popular categories in our sample. Users were restricted to
those who tweeted at least three BBC online URLs from the same category over the period
in which the tweets were collected. We obtained the list of “follow” relationships among the
set of users being studied.
Layout generation and clustering
The spatial layout of the network given in Figure 1 was determined solely by the topology
of the network – based on a force-directed layout algorithm, which optimizes the layout so
that nearby nodes push each other apart and edges pull the connected nodes closer [21, 25].
We applied the k-means unsupervised clustering algorithm on the two-dimensional coor-
dinates of the nodes to divide users into 7 different communities (clusters) [22].
Analysis of geographical locations
The profile of each Twitter user contains an optional location attribute as an unstructured
text field. We analyzed the text in this field to obtain the location of the users resulting in
the name of the city, the name of the country or unidentified text [26].
Analysis of overrepresented words
The public profile of each Twitter user contains a short biography of the user as a self-
provided text field. We analyzed this field to obtain information about the English speaking
12groups. In general, these fields contain non-standard language or typos. Nevertheless, it
is possible to capture the main characteristics of communities via aggregating text-based
statistics over a group.
We employed a bag-of-words technique for analyzing the profile descriptions [27]. We
lower-cased the texts, used the space character to split the text into words and did not
carry out any lemmatization – the surface forms of the words such as “works”, “working”
were counted separately. We removed any word containing characters outside the English
alphabet including é, è, and -. We also discarded stop words such as “and”, “the”, “of”
from the text.
A characterization of a cluster can be obtained by the most over-represented words in the
cluster. The bias of a word w, given two clusters c1 and c2 is b(w, c1 , c2 ) = p(w, c1 ) − p(w, c2 )
where p(w, c), the prevalence of a term w in a cluster c, is the ratio of users in c who mention
w in their bio fields.
[1] H. A. Simon, The Sciences of the Artificial, 3rd ed. (MIT Press, 1999).
[2] Y. Bar-Yam, Dynamics of Complex Systems (Westview Press, 1997).
[3] J. B. Thompson, The Media and Modernity: A Social Theory of the Media (Stanford Univer-
sity Press, Stanford, CA, USA, 1995).
[4] A. Briggs, P. Burke, Social History of the Media: From Gutenberg to the Internet (Polity
Press, 2002).
[5] A. Herdağdelen, W. Zuo, A. Gard-Murray, Y. Bar-Yam, An Exploration of Social Iden-
tity: The Geography and Politics of News-Sharing Communities in Twitter, arXiv:1202.4393
(2012).
[6] J. U. Henrikson, The growth of social media: An infographic, Search Engine Journal (2011).
[7] B. A. Nardi, D. J. Schiano, M. Gumbrecht, L. Swartz, Why we blog, Commununications of
the ACM , 47, 41–46 (2004).
[8] M. T. Poe, A History of Communications: Media and Society from the Evolution of Speech to
the Internet (Cambrigde University Press, 2011).
[9] B. Winston, A History of Communications: Media and Society from the Evolution of Speech
to the Internet (Routledge, New York, 2000).
13[10] M. S. Granovetter, The strength of weak ties, American Journal of Sociology, 78, 1360–1380
(1973).
[11] T. Heverin, L. Zach, Microblogging for Crisis Communication: Examination of Twitter Use
in Response to a 2009 Violent Crisis in the Seattle-Tacoma, Washington, Area (ISCRAM,
2010).
[12] J. P. Bagrow, D. Wang, A.-L. Barabási, Collective Response of Human Populations to Large-
Scale Emergencies, PLoS ONE , 6, 17680 (2011).
[13] The BBC Executive’s Review and Assessment (The British Broadcasting Corporation, Lon-
don, United Kingdom, 2012).
[14] A. Briggs, The History of Broadcasting in the United Kingdom: Volume II: The Golden Age
of Wireless (Oxford University Press, 1995).
[15] M. Thompson, The World Service can survive these cuts (2011). http://www.telegraph.co.
uk/culture/tvandradio/bbc/8281797/The-World-Service-can-survive-these-cuts.
html.
[16] T. Shaw, Eden, Suez and the Mass Media: Propanganda and Persuasion During the Suez
Crisis (I. B. Tauris & Co. Ltd, 1996).
[17] Analysis: BBC’s Voice in Europe (2005). http://news.bbc.co.uk/2/hi/europe/4375652.
stm Accessed July 15, 2013.
[18] B. News, Años del servicio latinoamericano de la BBC (2000). http://www.bbc.co.uk/
spanish/specials/1721 bbcmundoanivers/index.shtml Accessed July 15, 2013.
[19] Twitter, Numbers (2011). http://blog.twitter.com/2011/03/numbers.html Accessed Oc-
tober 22nd, 2011.
[20] J. Akshay, S. Xiaodan, F. Tim, T. Belle, Why We Twitter: Understanding Microblogging Us-
age and Communities, Proceedings of the Joint 9th WEBKDD and 1st SNA-KDD Workshop
2007 (2007).
[21] S. Martin, W. M. Brown, R. Klavans, K. W. Boyack, Openord: an open-source toolbox for
large graph layout, Conference on Visualization and Data Analysis, 7868, 786806 (2011).
[22] T. Hastie, R. Tibshirani, J. Friedman, J. Franklin, The elements of statistical learning: data
mining, inference and prediction (Springer Verlag, 2011).
[23] BBC World Service in Government Funding Cut (June 2013). http://www.bbc.co.uk/news/
entertainment-arts-22853598 Accessed July 15, 2013.
14[24] BBC World Service to Lose 73 Jobs (October 2012). http://www.bbc.co.uk/news/
entertainment-arts-19988316 Accessed July 15, 2013.
[25] M. Bastian, S. Heymann, M. Jacomy, Gephi: an open source software for exploring and
manipulating networks, International AAAI Conference on Weblogs and Social Media (2009).
[26] Geonames Web Services. http://www.geonames.org/ Accessed July 15, 2013.
[27] A. Kao, S. Poteet, Natural language processing and text mining (Springer Verlag, 2007).
15You can also read
